nemo_automodel.components.speculative.dflash.registry
nemo_automodel.components.speculative.dflash.registry
Dispatch registry mapping target architecture -> DFlash draft model.
Mirrors the EAGLE registry (components/speculative/eagle/registry.py). The
Qwen3 DFlash draft is a non-causal Qwen3-style stack and is config-driven, so
adding a Qwen3-shaped architecture is a one-line append; a target whose backbone
differs (Kimi K3’s MLA) registers its own draft class.
Module Contents
Classes
Functions
Data
API
Dataclass
How to build a DFlash draft model for a particular target architecture.
attention_backends
build_draft_config
build_target_kwargs
draft2_cls
draft_cls
supports_context_parallel
Default: a target needs no architecture-specific from_pretrained kwargs.
Return the first registered DFlash draft spec matching any architecture in the list.